1 /***************************************************************************
2                                 lj_gromacs.h
3                              -------------------
4                             Trung Dac Nguyen (ORNL)
5 
6   Class for acceleration of the lj/gromacs pair style.
7 
8  __________________________________________________________________________
9     This file is part of the LAMMPS Accelerator Library (LAMMPS_AL)
10  __________________________________________________________________________
11 
12     begin                :
13     email                : nguyentd@ornl.gov
14  ***************************************************************************/
15 
16 #ifndef LAL_LJ_GROMACS_H
17 #define LAL_LJ_GROMACS_H
18 
19 #include "lal_base_atomic.h"
20 
21 namespace LAMMPS_AL {
22 
23 template <class numtyp, class acctyp>
24 class LJGROMACS : public BaseAtomic<numtyp, acctyp> {
25  public:
26   LJGROMACS();
27   ~LJGROMACS();
28 
29   /// Clear any previous data and set up for a new LAMMPS run
30   /** \param max_nbors initial number of rows in the neighbor matrix
31     * \param cell_size cutoff + skin
32     * \param gpu_split fraction of particles handled by device
33     *
34     * Returns:
35     * -  0 if successful
36     * - -1 if fix gpu not found
37     * - -3 if there is an out of memory error
38     * - -4 if the GPU library was not compiled for GPU
39     * - -5 Double precision is not supported on card **/
40   int init(const int ntypes, double **host_cutsq,
41            double **host_lj1, double **host_lj2, double **host_lj3,
42            double **host_lj4, double *host_special_lj,
43            const int nlocal, const int nall, const int max_nbors,
44            const int maxspecial, const double cell_size,
45            const double gpu_split, FILE *screen,
46            double **host_ljsw1, double **host_ljsw2, double **host_ljsw3,
47            double **host_ljsw4, double **host_ljsw5,
48            double **cut_inner, double **cut_inner_sq);
49 
50   /// Clear all host and device data
51   /** \note This is called at the beginning of the init() routine **/
52   void clear();
53 
54   /// Returns memory usage on device per atom
55   int bytes_per_atom(const int max_nbors) const;
56 
57   /// Total host memory used by library for pair style
58   double host_memory_usage() const;
59 
60   // --------------------------- TYPE DATA --------------------------
61 
62   /// lj1.x = lj1, lj1.y = lj2, lj1.z = cutsq, lj1.w = cut_inner_sq
63   UCL_D_Vec<numtyp4> lj1;
64   /// lj3.x = lj3, lj3.y = lj4, lj3.z = cut_inner, lj3.w = ljsw5
65   UCL_D_Vec<numtyp4> lj3;
66   /// ljsw.x = ljsw1, ljsw.y = ljsw2, ljsw.z = ljsw3, ljsw.w = ljsw4
67   UCL_D_Vec<numtyp4> ljsw;
68   /// Special LJ values [0-3] and Special Coul values [4-7]
69   UCL_D_Vec<numtyp> sp_lj;
70 
71   /// If atom type constants fit in shared memory, use fast kernels
72   bool shared_types;
73 
74   /// Number of atom types
75   int _lj_types;
76 
77  private:
78   bool _allocated;
79   int loop(const int eflag, const int vflag);
80 };
81 
82 }
83 
84 #endif
85